The present invention relates to apparatus and methods for encoding picture signals, and, more particularly, to encoding of picture signals using predictive encoding and orthogonal transformations.
In a teleconferencing system or a video telephone system, for example, moving picture signals are compressed and encoded by taking advantage of intra-frame and inter-frame correlation so that they can be more efficiently transmitted over a communication channel to a remote location.
Intra-frame correlation can be utilized by an orthogonal transformation, such as a discrete cosine transformation (DCT).
Inter-frame correlation can be utilized by predictive encoding between successive pictures. As used herein, a picture generally refers to an image represented by a frame of video information. When the fields of a frame are coded in a non-interlaced manner, that is, separately, each field may be referred to as a picture.
As shown in FIG. 1A, for example, frame pictures PC1, PC2 and PC3 are generated at time points t1, t2 and t3. As shown by shading in FIG. 1B, the difference between the frame pictures PC1 and PC2 is obtained as difference picture data PC12, and the difference between the frame pictures PC2 and PC3 is obtained as difference picture data PC23. Since there is a fairly small change between signals of temporally neighboring frames, transmission of only the difference picture data utilizes the transmission channel more efficiently than transmission of the original pictures. That is, using the difference picture data as encoded picture signals, referred to as predictive encoding, reduces the amount of data to be transmitted.
However, if only the difference signals are transmitted, the original picture cannot be restored. It is advantageous to occasionally transmit a picture which is not predictively encoded as a reference for difference picture data, and because it is sometimes more efficient than transmitting the picture as a predictively encoded picture.
Pictures which are encoded utilizing only intra-frame correlation and not inter-frame correlation, are referred to herein as intra-pictures or I-pictures.
Pictures which are encoded with predictive encoding relative to one previously encoded picture are referred to herein as predictive pictures or P-pictures. The previously encoded picture may be an I-picture or a P-picture, and temporally precedes the P-picture.
Pictures which are encoded with predictive encoding relative to at most two pictures, that is, relative to only a temporally succeeding picture or relative to both a temporally preceding and a temporally succeeding picture, are referred to herein as bi-directionally predictive coded pictures or B-pictures. When the predictive encoding is relative to two pictures, the two pictures may be I-pictures or P-pictures, or one of each, and the mean value of the two pictures is obtained and used as a reference picture for the picture to be encoded.
A series of pictures may be considered as groups of pictures having a predetermined number of frames such as F1 . . . F17. The luminance and chrominance picture signals of the leading frame F1 are encoded as an I-picture, the picture signals of the second frame F2 are encoded as a B-picture, and the picture signals of the third frame F3 are encoded as a P-picture. The fourth and the following frames F4 to F17 are encoded alternately as B-pictures and P-pictures. FIG. 2A shows the reference pictures used for encoding P-pictures, while FIG. 2B shows the reference pictures used for encoding B-pictures.
As shown in FIGS. 3A and 3B, there are four methods for encoding the macro-blocks (discussed below) of a picture. When multiple methods are suitable, the method which will give the smallest amount of encoded data is employed on a macro-block by macro-block basis. Blocks F0 to F5 in FIG. 3A represent data for frames of moving picture signals, whereas blocks F1X to F5X in FIG. 3B represent data for encoded frames. The solid line arrows in FIG. 3A show the frames to which motion vectors x0 . . . x6 relate.
The first method, shown as SP1, is to not use predictive encoding, that is, to use only intra-frame correlation. This method is suitable for any macro-block of an I-picture, a P-picture and a B-picture. In other words, if less encoded data is produced without predictive encoding, then this method is selected.
The second method, shown as SP2, is to predictively encode relative to a picture which temporally succeeds the current picture, referred to as backward predictive encoding. The third method, shown as SP3, is to predictively encode relative to a picture which temporally precedes the current picture, referred to as forward predictive encoding. The second method is suitable for macro-blocks of only B-pictures. The third method is suitable for macros-blocks of P-pictures or B-pictures.
The fourth method, shown as SP4, is to predictively encode relative to the mean value of two pictures, one temporally preceding and one temporally succeeding the current picture. This method is suitable for macro-blocks of only B-pictures.
To encode a sequence of frames, frame F1 is first encoded as an I-picture using the first method SP1 so that it is directly transmitted over a transmission channel as encoded data F1X.
Frame F3 is then encoded as a P-picture. When the third method SP3, forward predictive coding, is used for a macro-block, a motion vector x3 representing the motion between the current picture F3 and the temporally preceding frame F1, used as the reference picture, and difference signals representing the difference between a motion compensated reference picture F1 and the current picture F3, as indicated by a broken-line arrow SP3, are calculated and encoded as data F3X for that macro-block. Alternatively, if a smaller amount of encoded data is produced for a macro-block of the P picture being encoded, the first method SP1 can be used wherein the data of the original frame F3 are directly utilized as the transmission data F3X for that macro-block.
Frame F2 is then encoded as a B-picture. When the fourth method SP4 is used to encode a macro-block of the frame F2, a difference between the mean value of the temporally preceding frame F1 and the temporally succeeding frame F3 is calculated, on a pixel by pixel basis. The difference data and the motion vectors x1 and x2 are encoded as data F2X. Alternatively, when the first processing method SP1 is used to encode a macro-block of the frame F2, the data of the original frame F2 forms the encoded data F2X.
When one of the second or third methods SP2, SP3 is used to encode a macro-block of the frame F2, one of the difference between the temporally succeeding frame F3 and the current frame F2, and the difference between the temporally preceding frame F1 and the current frame F2 is calculated. The difference data and one of the motion vectors x1, x2 are encoded as the data F2X.
The frame F4 for the B-picture and the frame F5 for the P-picture are processed in a similar manner as described above to generate transmitted data F4X and F5X.
Spatial redundancy within a picture is reduced by an orthogonal transformation, such as a discrete cosine transformation (DCT), of a portion of the picture from the time domain into the frequency domain. A block of pixel data from the picture having a dimension of, for example, 8 pixels width .times.8 rows height, representing luminance or chrominance amplitudes at the respective pixels, is converted by DCT transformation into a block of 8.times.8 frequency coefficients, which is scanned in a predetermined zigzag manner from low frequency to high frequency to provide a sequence of 64 coefficients representing the amounts of respective frequencies contained in the block. The first coefficient is referred to as the DC coefficient, while the other 63 coefficients are referred to as the AC or high frequency coefficients. A pixel block representing a solid portion of an image corresponds to a DC coefficient indicating the amplitude of the solid portion, and no high frequency coefficients. A pixel block representing a highly detailed image portion corresponds to coefficient data with many non-zero AC values.
A picture of a natural scene tends to be smooth, that is, to lack highly detailed image portions. Consequently, the pixel blocks of such a picture correspond to DCT coefficients lacking AC coefficients, that is, having runs of zero data. These runs of zero data are variable length coded by representing them as a run-length number indicating how many zeros are in the run. The run-length value is further encoded using a Huffman code.
At the decoder, the encoded signal is variable length decoded (inverse variable length coded), and then inverse DCT transformed to recover the original pixel data for the picture.
Applying the techniques of predictive encoding and orthogonal transformation to a picture sequence removes significant amounts of temporal and spatial redundancy from the picture sequence and results in a highly efficiently encoded representation of the picture sequence.
An area of a picture includes pixels in odd-numbered rows and pixels in even-numbered rows. In progressive scanning, the rows of pixels in a frame are scanned from top to bottom. During display, the pixels are presented in this same order. In interlace scanning, first the odd-numbered rows of pixels forming an odd field in a frame are scanned, then the even-numbered rows of pixels forming an even field in the frame are scanned. During display, the odd field is displayed and then the even field is displayed such that its rows are interlaced with the rows in the odd field.
If motion is represented in a sequence of interlaced scanned pictures, each frame exhibits comb-like deformation. FIG. 4 shows an image of a car moving from the left side of the picture to the right side of the picture. When the odd field is scanned, the car is in one position. By the time that the even field is scanned, the car has advanced towards the right. During display of the interlaced scanned fields of a frame representing this picture, the edges represented by the even field are shifted with respect to the edges represented by the odd field, causing the edges of an object to appear jagged. The comb deformation may be particularly seen in a vertical edge, such as the front of the car.
In frame-based processing of an interlaced scanned picture, the pixel rows in an area of the picture are processed while the odd-numbered rows and even-numbered rows are interlaced, as shown in FIG. 5A. DCT transformation is typically performed on a block of interlaced picture data of size 8 pixels .times.8 rows.
In field-based processing of an interlaced scanned picture, the rows in the odd field in an area of a picture are separated from the rows in the even field in that area, and all of the rows in the odd field are processed separately from the rows in the even field, as shown in FIG. 5B. DCT transformation is typically performed on an 8.times.8 block of data representing either an odd field or an even field of a picture area, that is, DCT transformation is performed on an 8.times.8 block of segregated picture data.
The amount of DCT coefficient data needed to represent a picture is influenced by the nature of the image represented in the picture. As an example, the amount of motion represented in a picture often has a large influence on the orthogonal transformation. As shown in FIG. 4, an area representing the street is uniform and is suitable for frame-based DCT transformation. However, an area representing the front edge of the car in FIG. 4 is ill-suited for frame-based DCT transformation, as a substantial amount of high frequency coefficients in a transformed block will have non-zero value, reducing the length of zero runs, and thus reducing the efficiency of the variable length code.
In contrast, in field-based processing, the rows in the odd field in the edge of the car are considered separately from the rows in the even field so that no abrupt transitions need be represented, that is, only smoothly changing edges need be represented, dramatically reducing the amount of high frequency coefficient data relative to when the rows are processed in interlaced form. Correspondingly, the zero runs in field-based DCT transformation of the edge of the car are longer than those in frame-based DCT transformation, so less variable length code encoded data is needed to represent the car edge using field-based DCT transformation.
FIG. 6 illustrates an arrangement for encoding and decoding moving picture signals in accordance with the above-described predictive encoding scheme. As shown in FIG. 6, a coding device 1 encodes input picture signals and transmits the encoded signals to a recording medium 3 as a transmission channel for recording. A decoding device 2 reproduces the signals recorded on the recording medium 3 and decodes these as output signals.
The coding device 1 includes an input terminal 10, a pre-processing circuit 11, A/D converters 12 and 13, a frame memory 14 including a luminance signal frame memory 15 and a color difference signal frame memory 16, a format converter 17 and an encoder 18.
Input terminal 10 is adapted to receive a video signal VD and to supply the signal VD to pre-processing circuit 11 which functions to separate the video signal VD into luminance signals and color signals, herein chrominance or color difference signals, that are applied to analog-to-digital (A/D) converters 12 and 13, respectively. The video signals, digitized by analog-to-digital conversion by the A/D converters 12 and 13, are supplied to frame memory 14 having memories 15, 16 which function to store the luminance signals and the color difference signals, respectively, and to read out the signals stored therein to format converter 17.
The converter 17 is operative to convert frame format signals stored in the frame memory section 14 into block format signals. As shown in FIG. 7A, pictures are stored in the frame memory section 14 as frame-format data having V lines each consisting of H dots. The converting circuit 17 divides each frame into N slices, each slice comprising a multiple of 16 lines. As shown, in FIG. 7B, the converter 17 divides each slice into M macro-blocks. As shown in FIG. 7C, each macro-block represents luminance signals Y corresponding to 16.times.16 pixels or dots, and associated chrominance Cr, Cb signals. These luminance signals are sub-divided into blocks Y1 to Y4, each consisting of 8.times.8 dots. The 16.times.16 dot luminance signals are associated with 8.times.8 dot Cb signals and 8.times.8 dot Cr signals. The converter 17 is also operative to supply the block format signals to the encoder 18, which is described in detail below with reference to FIG. 8.
The encoder 18 operates to encode the block format signals and to supply the encoded signals as a bitstream over a transmission channel for recording on the recording medium 3.
The decoding device 2 includes a decoder 31, a format converter 32, a frame memory section 33 including a luminance signal frame memory 34 and a color difference signal frame memory 35, digital-to-analog converters 36 and 37, a post-processing circuit 38 and an output terminal 30.
The decoder 31 is operative to reproduce encoded data from the recording medium 3 and to decode the encoded data, as described in detail below with reference to FIG. 12, and to supply decoded data signals to format converter 32 which is operative to convert the decoded data signals into frame format data signals and to supply the frame format data signals as luminance signals and color difference signals to the memory 33. The memories 34, 35 of the memory 33 function to store the luminance and chrominance signals, respectively, and to apply these signals to D/A converters 36 and 37, respectively. The analog signals from converters 36, 37 are synthesized by a post-processing circuit 38 which functions to form output picture signals and to output them to output terminal 30, and thence to a display unit, such as a CRT, not shown, for display.
FIG. 8 illustrates the encoder 18 shown in FIG. 6. Generally, the encoder 18 stores three pictures, the current picture and the pictures temporally preceding and succeeding the current picture. Based on the sequential position of the current picture in the group of pictures, the picture coding type (I, P or B) is selected for each picture.
The encoder 18 also chooses one of frame-based and field-based predictive encoding as will be explained with reference to FIG. 9, and further chooses one of frame-based and field-based DCT encoding as will be explained with reference to FIG. 10. For each picture, appropriate motion vectors are obtained and the picture is predictively encoded relative to zero, one or two previously encoded pictures which have been locally decoded and which are referred to as reference pictures to form a difference data signal. The difference data signal is orthogonally transformed into blocks of coefficient data which are quantized, variable length encoded and transmitted as encoded data.
The encoder 18 includes a local decoder in which the quantized data are dequantized, inverse orthogonally transformed, and stored as the reference pictures. In accordance with predictive encoding, the motion vector(s) obtained for the current picture is (are) applied to the reference picture(s) to produce a predictive picture which is subtracted from the current picture to yield the difference data.
The elements of the encoder 18 will now be explained in detail.
Picture data for encoding is supplied macro-block by macro-block to an input terminal 49 and thence to a motion vector detector 50 which is operative to process the picture data of respective frames as I-pictures, P-pictures or as B-pictures, in accordance with a predetermined sequence for each group of pictures, as shown for example, in FIGS. 2A, 2B. The circuit 50 applies the picture data of the current frame to a frame memory 51 having frame memories 51a, 51b, 51c used for storing a temporally preceding picture, the current picture and a temporally succeeding picture, respectively.
More specifically, the frames F1, F2, F3 are stored in the memories 51a, 51b, 51c, respectively. Then the picture stored in memory 51c is transferred to memory 51a and the frames F4, F5 are stored in the memories 51b, 51c, respectively. The operations of transferring the picture in memory 51c to memory 51a and storing the next two pictures in memories 51b, 51c are repeated for the remaining pictures in the group of pictures.
After the pictures are read into the memory and temporarily stored, they are read out and supplied to a predictive mode switching circuit 52 which is adapted to process the current picture for one of frame-based and field-based predictive encoding. After processing the first frame picture data in a group of pictures as an I-picture and before processing the second frame picture as a B-picture, the motion vector detector 50 processes the third frame P-picture. The processing sequence is different from the sequence in which the pictures are supplied because the B-picture may involve backward prediction, so subsequent decoding may require that the P-picture temporally succeeding the B-picture have been previously decoded.
The motion vector detector 50 calculates as an estimated value for intra-coding for each macro-block, the sum of absolute values of predictive errors for the frame predictive mode for each macro-block and the sum of absolute values of predictive errors for the field predictive mode for each macro-block and supplies these sums to the predictive judging circuit 54 which compares these sums and selects frame predictive mode or field predictive mode in accordance with the smallest of these values and provides the selected mode to predictive mode switching circuit 52.
If the frame predictive mode is selected, the predictive mode switching circuit 52 outputs the four luminance blocks Y1 to Y4 and the two chrominance or color difference blocks Cb, Cr of each macro-block received from the motion vector detector 50 without processing. As shown in FIG. 9A, odd or first field line data, indicated by solid lines, and even or second field line data, indicated by dashed lines, alternate in each luminance and color difference block as received from the motion vector detector 50. In FIG. 9A, a indicates units for motion compensation. In the frame predictive mode, motion compensation is performed with four luminance blocks (macro-blocks) as a unit and a single motion vector is associated with the four luminance blocks Y1 to Y4.
If the field predictive mode is selected, the predictive mode switching circuit 52 processes the signals received from the motion vector detector 50 so that each of the four luminance blocks comprises data from a single field and the two color difference blocks have non-interlaced odd and even field data. Specifically, as shown in FIG. 9B, the luminance blocks Y1 and Y2 have odd-field data and the luminance blocks Y3 and Y4 have even-field data, while the upper halves of the color difference blocks Cb, Cr represent odd field color difference data for the luminance blocks Y1 and Y2 and the lower halves of the color difference blocks Cb, Cr represent even field color difference data for the luminance blocks Y3 and Y4. In FIG. 9B, b indicates units for motion compensation. In the field predictive mode, motion compensation is performed separately for the odd-field blocks and even-field blocks so that one motion vector is associated with the two luminance blocks Y1 and Y2 and another motion vector is associated with the two luminance blocks Y3 and Y4.
The predictive mode switching circuit 52 supplies the current picture, as processed for frame-based or field-based predictive encoding, to arithmetic unit 53 of FIG. 8. The arithmetic unit 53 functions to perform one of intra-picture prediction, forward prediction, backward prediction or bi-directional prediction. A predictive judging circuit 54 is adapted to select the best type of prediction in dependence upon the predictive error signals associated with the current picture signals.
The motion vector detector 50 calculates, for the current picture, the sum of the absolute values of the differences between each signal Aij and the average value A of the signals Aij in each macro-block, .SIGMA..vertline.Aij-A.vertline., and supplies the sum as an estimated value for intra-coding to the predictive judging circuit 54.
The motion vector detector 50 calculates the sum of absolute values (or sum of squares) of the difference (Aij-Bij) between signals Aij of the macro-blocks of the current picture, and signals Bij of the macro-blocks of the predictive picture, .SIGMA..vertline.Aij-Bij.vertline., in each of the frame predictive mode and the field predictive mode. As explained above, the motion vector(s) for the current picture are applied to the reference picture(s) to generate the predictive picture. When the reference picture temporally precedes the current picture, the quantity .SIGMA..vertline.Aij-Bij.vertline. is referred to as a forward predictive error signal, and when the reference picture temporally succeeds the current picture, the quantity .SIGMA..vertline.Aij-Bij.vertline. is referred to as a backward predictive error signal. When the predictive picture is the mean of a temporally preceding and a temporally succeeding reference picture, as motion-compensated, the quantity .SIGMA..vertline.Aij-Bij.vertline. is referred to as a bi-directional predictive error signal.
The motion vector detector 50 supplies the forward frame predictive, the forward field predictive, the backward frame predictive, the backward field predictive, the bi-directional frame predictive and the bi-directional field predictive error signals to the predictive judging circuit 54.
The predictive judging circuit 54 selects one of intra-coding, forward inter-picture prediction, backward inter-picture prediction or bi-directional inter-picture prediction and one of the frame and field predictive modes in accordance with the smallest of the estimated value for intra-coding and the forward frame, the forward field, the backward frame, the backward field, the bi-directional frame and the bi-directional field predictive error signals. The arithmetic unit 53 predictively encodes the current picture, as processed by the predictive mode switching circuit 52, in accordance with the predictive mode selected by the predictive judging circuit 54.
The motion vector detector 50 serves to calculate and supply the motion vector(s) associated with the selected predictive mode to a variable length coding circuit 58 and a motion compensation circuit 64, explained later.
The sums of the absolute values of the inter-frame differences (predictive errors) on the macro-block basis are supplied from the motion vector detector 50 to the predictive mode switching circuit 52 and to the predictive judging circuit 54, in the manner as described above.
The arithmetic unit 53 supplies predictively encoded data, also referred to as difference data, for the current picture to a DCT mode switching circuit 55 which is adapted to process the current picture for one of frame-based and field-based orthogonal transformation.
The DCT switching circuit 55 functions to compare the encoding efficiency when the DCT operations for the macro-blocks in a picture are performed with the odd field data alternating with the even field data, that is, for frame-based orthogonal transformation, as shown in FIG. 10A, with the encoding efficiency when the DCT operations for the macro-blocks in a picture are performed with the odd field data separated from the even field data, that is, for field-based orthogonal transformation, as shown in FIG. 10B. The circuit 55 functions to select the mode with the higher encoding efficiency.
To evaluate the encoding efficiency for frame-based orthogonal transformation, the DCT mode switching circuit 55 places the luminance macro-block data into interlaced form, as shown in FIG. 10A, and calculates the differences between the odd field line signals and even field line signals vertically adjacent to each other, and finds the sum of absolute values of the differences EFM, or the sum of squared values of the differences. ##EQU1##
To evaluate the encoding efficiency for field-based orthogonal transformation, the DCT mode switching circuit 55 places the luminance macro-block data into non-interlaced form, as shown in FIG. 10B, and calculates the differences between vertically adjacent odd field line signals and the differences between vertically adjacent even field line signals, and finds the sum of absolute values of the differences EFD, or the sum of squared values of the differences. ##EQU2##
The DCT switching circuit 55 compares the difference between the frame-based and field-based sums of the absolute values with a predetermined threshold and selects frame-based DCT transformation if the difference EFM-EFD is less than the predetermined threshold.
If the frame predictive mode is selected in the predictive judging circuit 54, the probability is high that the frame DCT mode will be selected in the DCT mode switching circuit 55. If the field predictive mode is selected in the predictive judging circuit 54, the probability is high that the field DCT mode will be selected in the DCT mode switching circuit 55. However, since this is not necessarily the case, the predictive judging circuit 54 sets the mode which will give the least value of the sum of the absolute values of predictive errors, while the DCT mode switching circuit 55 sets the mode which will give the optimum orthogonal transformation encoding efficiency.
If frame-based orthogonal transformation mode, also referred to as frame DCT mode, is selected, the DCT mode switching circuit 55 functions to ensure that the four luminance blocks Y1 to Y4 and two color difference blocks Cb, Cr represent alternating or interlaced odd and even field lines, as shown in FIG. 10A.
If a field-based orthogonal transformation mode, also referred to as field DCT mode, is selected, the DCT mode switching circuit 55 functions to ensure that each of the luminance blocks represents only one field, and that each of the color difference blocks has segregated or non-interlaced odd and even field lines, as shown in FIG. 10B.
The DCT mode switching circuit 55 functions to output the data having the configuration associated with the selected DCT mode, and to output a DCT flag indicating the selected DCT mode to the variable length coding circuit 58 and the motion compensation circuit 64.
FIG. 11 shows an example of a configuration of the DCT mode switching circuit 55. The subtractor elements 160 calculate the differences between respective pixels of vertically adjacent odd and even field line signals of interlaced macro-block (a). The squaring circuits 161 square the respective differences and supply the squared respective differences to the adder 162 where the sum of the squared differences, EFM, is determined. The adder 162 supplies the sum EFM to the comparator 165.
Similarly, the subtractors 170 calculate the differences between pixels of vertically adjacent odd field line signals and calculate the differences between pixels of vertically adjacent even field line signals of the macro-block, as shown in a field-based configuration (b). The squaring circuits 171 square the respective differences and the adder 172 calculates the sum of the squared differences, EFM, and supplies the sum EFM to the comparator 165.
The comparator 165 compares the difference between the sum EFD and the sum EFM with a predetermined threshold value and selects one from the frame DCT mode or the field DCT mode accordingly.
Referring back to FIG. 8, the DCT mode switching circuit 55 supplies appropriately configured difference picture data to a DCT circuit 56 which is operative to orthogonally transform it using a discrete cosine transformation into DCT coefficients, and to supply the DCT coefficient data to a quantization circuit (quantizer) 57 that functions to quantize the coefficient data with quantization steps selected in accordance with the volume of data stored in a transmission buffer 59 and to supply quantized data to a variable length coding circuit 58.
The variable length coding circuit 58 is also supplied with the quantization step or scale data from the quantizer 57, predictive mode data from the predictive judging circuit 54, that is data indicating which of the intra-picture prediction, forward prediction, backward prediction or bi-directional prediction is used, and motion vector data from the motion vector detector 50. The coding circuit 58 also receives prediction flag data from the predictive judging circuit 54 comprising a flag indicating which of the frame predictive mode or the field predictive mode is used, and prediction flag data from the DCT mode switching circuit 55 comprising a flag indicating which of the frame DCT mode or the field DCT mode is used. This information is placed into the header portion of the encoded data stream.
The variable length coding circuit 58 serves to encode the quantized data and the header information using a variable length code such as a Huffman code, in accordance with the quantization step data supplied from the quantizer 57, and to output the resulting data to the transmission buffer 59.
The quantized data and quantization step are also supplied to an inverse quantization (inverse quantizer) circuit 60 which serves to dequantize the quantized data using the quantization step, and to supply the recovered DCT coefficient data to an inverse DCT circuit 61 that functions to inverse transform the DCT coefficient data to produce recovered difference data and to supply the recovered difference data to an arithmetic unit 62.
The arithmetic unit 62 combines the recovered difference data with a previously encoded and decoded reference picture, as motion compensated, to produce decoded data for a reconstructed picture which will be used as a reference picture and which is read into one of two frame memories 63a, 63b. The memories 63a, 63b are adapted to read out the reference picture data stored therein to a motion compensation circuit 64 that uses the motion vectors from the motion vector detector 50 to produce a predictive picture from the reference picture. Specifically, the circuit 64 uses the motion vector to alter the readout address of the reference picture from the memory 63a or 63b.
For a group of pictures, after the first frame I-picture data and the third frame P-picture data are stored in the forward and backward prediction picture memories or units 63a, 63b, respectively, the second frame B-picture data is processed by the motion vector detector 50. The predictive judging circuit 54 selects the frame or field predictive mode, while setting the predictive mode to one of intra-frame predictive mode, forward predictive mode, backward predictive mode or bi-directional predictive mode in correspondence with the sum of absolute values of predictive errors by macro-block.
Since a reconstructed B-picture is not used as a reference picture for other pictures, it is not stored in the frame memory 63.
It will be appreciated that the frame memory 63 has its forward and backward predictive picture units 63a, 63b bank-exchanged as needed so that a picture stored in one of the units 63a or 63b can be outputted as either a forward or a backward predictive picture.
The motion compensation circuit 64 functions to supply the motion compensated data as a predictive picture to the arithmetic unit 62 and to the arithmetic unit 53 which subtracts the predictive picture from the P-picture or the B-picture currently being predictively encoded.
More specifically, when the motion vector detector 50 receives picture data for an I-picture from the forward original picture unit 51a, the predictive judging circuit 54 selects the intra-frame predictive mode and sets a switch 53d of the arithmetic unit 53 to an input contact a. This causes the I-picture data to be inputted directly to the DCT mode switching circuit 55. In this case, no predictive picture is expected from the motion compensation circuit 64. The I-picture data is also supplied to the forward predictive picture memory unit 63a.
When the forward predictive mode is selected by the predictive judging circuit 54, the circuit 54 sets the switch 53d to an input contact b which causes the arithmetic unit 53a to subtract the predictive picture, produced by the motion compensation circuit 64, from the picture read out from the memory 51, for each macro-block on a pixel by pixel basis, to produce difference data. The P-picture, after encoding and local decoding, is supplied to one of the memory units 63a, 63b. For example, if the P-picture immediately follows an I-picture, then the P-picture is stored in the backward predictive picture memory unit 63b.
For forward predictive encoding, the predictive picture is a reference I-picture or P-picture read out from the forward predictive picture unit 63a of the frame memory 63 and motion-compensated by the motion compensation circuit 64 in accordance with the motion vector outputted from the motion vector detector 50. More specifically, for each macro-block, the motion compensation circuit 64 shifts the readout address of the forward predictive picture memory unit 63a in an amount corresponding to the motion vector currently output by the motion vector detector 50.
When the backward predictive mode is selected by the predictive judging circuit 54, the circuit 54 sets the switch 53d to an input contact c which causes the arithmetic unit 53b to subtract the predictive picture, produced by the motion compensation circuit 64, from the picture read out from the memory 51, on a pixel by pixel basis, to produce difference data.
For backward predictive encoding, the predictive picture is a P-picture read out from the backward predictive picture unit 63b of the frame memory 63 and motion-compensated by the motion compensation circuit 64 in accordance with the motion vector outputted from the motion vector detector 50. More specifically, for each macro-block, the motion compensation circuit 64 shifts the readout address of the backward predictive picture memory unit 63b in an amount corresponding to the motion vector currently output by the motion vector detector 50.
When the bi-directional predictive mode is selected by the predictive judging circuit 54, the circuit 54 sets the switch 53d to an input contact d which causes the arithmetic unit 53c to subtract a predictive picture from the picture read out from the memory 51, on a pixel by pixel basis, to produce difference data. The predictive picture is the mean value of a forward predictive picture and a backward predictive picture.
In the case of bi-directional prediction, the picture stored in the forward predictive picture memory unit 63a, and the picture stored in the backward predictive picture memory unit 63b, are read out and motion-compensated by the motion compensation circuit 64 in dependence upon the motion vectors outputted from the motion vector detector 50. More specifically, for each macro-block, the motion compensation circuit 64 shifts the readout address of the forward and backward predictive picture memory units 63a, 63b in an amount corresponding to the appropriate one of the motion vectors currently output by the motion vector detector 50.
The transmission buffer 59 temporarily stores the data supplied thereto, generates control data indicating the volume of data stored therein and supplies the control data to the quantizer 57. When the volume of data stored in the transmission buffer 59 reaches a predetermined upper limit value, the control data from the transmission buffer 59 causes the quantization scale of the quantizer 57 to increase so as to decrease the volume of the quantized data. Similarly, when the volume of data stored in the transmission buffer 59 reaches a predetermined lower limit value, the control data from the transmission buffer 59 causes the quantization scale of the quantizer 57 to decrease so as to increase the volume of the quantized data. In this manner, the transmission buffer 59 prevents the data supplied thereto from overflowing or underflowing its capacity. The data stored in the transmission buffer 59 are read out at a predetermined timing to an output terminal 69 and thence to a transmission channel for recording on, for example, the recording medium 3.
Although the foregoing description has been made with reference mainly to the luminance blocks, the color difference blocks are similarly processed and transmitted using the motion vector which corresponds to the motion vector of the luminance block halved in both the vertical and horizontal directions.
FIG. 12 illustrates the decoder 31 shown in FIG. 6. The reproduced encoded picture data transmitted from the recording medium 3 is applied to a reception circuit, not shown, or to an input terminal 80 which applies the encoded picture data to a receiving buffer 81 that serves to temporarily store the encoded picture data and to supply this data to a variable length decoding circuit 82 of a decoding circuit 90.
The variable length decoding circuit 82 functions to variable length decode the encoded data, to output the recovered motion vector, predictive mode data, prediction flags and DCT flags to the motion compensation circuit 87, and to output the quantization step data and variable length decoded picture data, including the predictive mode, the motion vector, the predictive flag, the DCT flag and the quantized picture data for each macro-block, to an inverse quantizer 83.
The inverse quantizer 83 is adapted to dequantize the picture data supplied from the variable length decoding circuit 82 in accordance with the quantization step data supplied from the variable length decoding circuit 82 and to output the thus recovered coefficient data to an inverse transformation IDCT circuit 84.
The IDCT circuit 84 is adapted to perform an inverse transformation on the recovered coefficient data to produce recovered difference data, and to supply the recovered difference data to an arithmetic unit 85.
If the recovered difference data supplied from the IDCT circuit 84 represents a macro-block of an I-picture, the arithmetic unit 85 does not process the data and simply supplies it through an output terminal 91 to the format converter 32 shown in FIG. 6, and to a forward predictive picture unit 86a of a frame memory 86.
If the recovered difference data supplied from the IDCT circuit 84 represents a macro-block of a P-picture produced in the forward predictive mode, then the reference picture data of the preceding frame, as stored in the forward predictive picture memory 86a of the frame memory 86, is read and motion-compensated by a motion compensation circuit 87 in dependence upon the motion vector outputted from the variable length decoding circuit 82 to generate a prediction picture. Specifically, the motion compensation circuit 87 uses the motion vector to alter the read out address supplied to the memory 86a. The arithmetic unit 85 adds the predictive picture to the recovered difference data to produce a decoded or reconstructed picture which is stored in a backward predictive picture memory 86b of the frame memory 86. The decoded P-picture is retained in the decoder 31, and output after the next B-picture is decoded and output, so as to restore the pictures to the order in which they were supplied to the encoder 18 of FIG. 6.
Even if the macro-block of the P-picture was encoded as intra-coded data, the decoded P-picture is directly stored in the backward predictive picture memory unit 86b, without being output to the output terminal 91 by the arithmetic unit 85.
If the recovered difference data supplied from the IDCT circuit 84 represents a macro-block of a B-picture encoded in the intra-coding mode, as determined from the predictive mode supplied from the variable length decoding circuit 82 to the motion compensation circuit 87, a predictive picture is not generated therefrom.
If the recovered difference data supplied from the IDCT circuit 84 represents a macro-block of a B-picture encoded in the forward predictive mode, as determined from the predictive mode supplied from the variable length decoding circuit 82 to the motion compensation circuit 87, the data stored in the forward predictive picture unit 86a of the frame memory 86 is read out and motion compensated by the motion compensation circuit 87 using the motion vector supplied from the variable length decoding circuit 82 to form the predictive picture. The arithmetic unit 85 sums the recovered difference data with the predictive picture to form the recovered B-picture.
If the recovered difference data supplied from the IDCT circuit 84 represents a macro-block of a B-picture encoded in the backward predictive mode, as determined from the predictive mode supplied from the variable length decoding circuit 82 to the motion compensation circuit 87, the data stored in the backward predictive picture unit 86b is read out and motion compensated by the motion compensation circuit 87 using the motion vector supplied from the variable length decoding circuit 82 to form the predictive picture. The arithmetic unit 85 sums the recovered difference data with the predictive picture to form the recovered B-picture.
If the recovered difference data supplied from the IDCT circuit 84 represents a macro-block of a B-picture encoded in the bi-directional predictive mode, as determined from the predictive mode supplied from the variable length decoding circuit 82 to the motion compensation circuit 87, the data stored in both the forward and backward predictive picture memories 86a, 86b are read out and respectively motion compensated by the motion compensation circuit 87 using the motion vectors supplied from the variable length decoding circuit 82, then averaged to form the predictive picture. The arithmetic unit 85 sums the recovered difference data with the predictive picture to form the recovered B-picture.
The recovered B-picture is supplied via the output terminal 91 to the format converter 32. However, since the B-picture is not utilized for generating a predictive picture for other pictures, it is not stored in the frame memory 86.
After outputting of the B-picture, picture data of the P-picture stored in the backward predictive picture unit 86b is read and supplied via the motion compensation circuit 87 to the arithmetic unit 85. Motion compensation is not performed at this time.
The counterpart circuits to the predictive mode switching circuit 52 and the DCT mode switching circuit 55 in the encoder 18 of FIG. 8 are not shown in the decoder 31. The processing to be performed by these circuits, that is, the processing for restoring the configuration in which odd-field line signals and even-field line signals are separated from each other to the configuration in which odd and even-field line signals alternate with each other, is performed by the motion compensation circuit 87.
The processing of the luminance signals has been explained in the foregoing. As will be appreciated by one of ordinary skill in the art, the processing of the color difference signals is carried out in a similar manner. However, the motion vector employed in such case is the motion vector for luminance signals which is halved in both the vertical and horizontal directions.
The conventional encoding and decoding methods for picture signals described above, have the drawback that the differences EFM and EFD, which are used to select either the frame-based DCT mode or the field-based DCT mode, do not necessarily predict the orthogonal transformation mode having the greater efficiency.
A further drawback of the above-described conventional encoding methods is that because selection from one of intra-picture prediction, forward inter-picture prediction, backward inter-picture prediction or bi-directional inter-picture prediction is made using predictive error signals that are not a function of the DCT coefficients, the choice of picture prediction does not necessarily result in the most efficient orthogonal transformation of the predictive picture.